LEARN. PRACTICE. GROW.Concept and mechanism
A defect metric is useful only when its population and observation time are known. If 40 defects were removed before release and ten identified afterward, removal efficiency over known defects is 40 divided by 50, or 80%. If another ten escaped defects are later found, the same calculation becomes 40 divided by 60, about 66.7%. The historical product did not change in this calculation; knowledge of the denominator changed. Always state the observation window and avoid comparing teams with different recording practices. Unknown defects do not enter the count, so the indicator does not prove the absence of remaining failures.
Guided application
To study containment by phase, connect where a defect was introduced with where it was removed. If a phase introduced 20 known defects and removed 15 within that phase, containment is 75%; the remaining five escaped into later phases. In iterative cycles, define coherent analysis units without imposing artificial phases on the work. A frequent category, such as data validation, identifies an investigation area but does not reveal cause by itself. Seek evidence about requirements, models, interfaces, and feedback that allowed the problem. Choose a concrete action, owner, and evaluation criterion. For example, reviewing boundaries with rejection examples may reduce ambiguity; check this through decision quality and recurrence without promising causality from a small sample.
40/(40+10)=80%; with ten more known defects,40/60≈66.7%.
Common pitfalls
Assumed fixed denominator; frequency treated as cause; comparing different periods.
Related topics: Analysis and evidence in the test process · Priority and residual risk · Domains, boundaries, and combinations
Define the metric and test the improvement hypothesis.
Reference: ISTQB CTAL-TA syllabus · CTAL-TA v4.0 (2025)